Merge pull request #1 from staskh/python3.11

Upgrade to PyTorch 2.3 and Python 3.11

.gitignore

@@ -127,3 +127,6 @@ dmypy.json
 
 # Pyre type checker
 .pyre/
+# do not save Results in Git
+Results/*
+M1_Retinal_Image_quality_EyePACS/test_outside/results_ensemble.csv

.vscode/launch.json

@@ -0,0 +1,27 @@
+{
+    // Use IntelliSense to learn about possible attributes.
+    // Hover to view descriptions of existing attributes.
+    // For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
+    "version": "0.2.0",
+    "configurations": [
+        {
+            "name": "Python Debugger: Current File with Arguments",
+            "type": "debugpy",
+            "request": "launch",
+            "program": "${file}",
+            "console": "integratedTerminal",
+            "cwd": "${fileDirname}",
+            "justMyCode": true,
+            "args": "${command:pickArgs}"
+        },
+        {
+            "name": "Debug",
+            "type": "debugpy",
+            "request": "launch",
+            "program": "${file}",
+            "console": "integratedTerminal",
+            "justMyCode": true,
+            "cwd": "${fileDirname}"
+        }
+    ]
+}

LOCAL.md

@@ -2,18 +2,18 @@
 
 ### Requirements
 
-1. Linux is preferred. For windows, install [MinGW-w64](https://www.mingw-w64.org/) for using commands below to set enviroment.
+1. Linux or Mac are preferred. For windows, install [MinGW-w64](https://www.mingw-w64.org/) for using commands below to set enviroment.
 2. Anaconda or miniconda installed.
-3. python=3.6, cudatoolkit=11.0, torch=1.7, etc. (installation steps below)
-4. GPU is essential.
+3. python=3.11, torch=2.3, etc. (installation steps below)
+4. GPU is essential -  NVIDIA (cuda) or M2 (mps).
 
 
 ### Package installation
 
 Step 1: create virtual environment:
 ```bash
 conda update conda
-conda create -n automorph python=3.6 -y
+conda create -n automorph python=3.11 -y
 ```
 
 Step 2: Activate virtual environment and clone the code.
@@ -23,16 +23,15 @@ git clone https://github.com/rmaphoh/AutoMorph.git
 cd AutoMorph
 ```
 
-Step 3: install pytorch 1.7 and cudatoolkit 11.0
+Step 3: install pytorch 2.3
 ```bash
-conda install pytorch==1.7.0 torchvision==0.8.0 torchaudio==0.7.0 cudatoolkit=11.0 -c pytorch -y
+conda install pytorch==2.3.1 -c pytorch -y
 ```
 
 Step 4: install other packages:
 ```bash
 pip install --ignore-installed certifi
 pip install -r requirement.txt
-pip install efficientnet_pytorch
 ```
 
 ### Running

M0_Preprocess/fundus_prep.py

@@ -72,7 +72,7 @@ def _get_radius_by_mask_center(mask,center):
     # radius=
     index=np.where(mask>0)
     d_int=np.sqrt((index[0]-center[0])**2+(index[1]-center[1])**2)
-    b_count=np.bincount(np.ceil(d_int).astype(np.int))
+    b_count=np.bincount(np.ceil(d_int).astype(int))
     radius=np.where(b_count>b_count.max()*0.995)[0].max()
     return radius
 
@@ -129,7 +129,7 @@ def mask_image(img,mask):
 def remove_back_area(img,bbox=None,border=None):
     image=img
     if border is None:
-        border=np.array((bbox[0],bbox[0]+bbox[2],bbox[1],bbox[1]+bbox[3],img.shape[0],img.shape[1]),dtype=np.int)
+        border=np.array((bbox[0],bbox[0]+bbox[2],bbox[1],bbox[1]+bbox[3],img.shape[0],img.shape[1]),dtype=int)
     image=image[border[0]:border[1],border[2]:border[3],...]
     return image,border
 

M1_Retinal_Image_quality_EyePACS/test_outside.py

@@ -12,19 +12,19 @@
 import pandas as pd
 import torch.nn as nn
 from tqdm import tqdm
-from pycm import *
-import matplotlib
-import matplotlib.pyplot as plt
+#from pycm import *
+# import matplotlib
+# import matplotlib.pyplot as plt
 from dataset import BasicDataset_OUT
 from torch.utils.data import DataLoader
 from model import Resnet101_fl, InceptionV3_fl, Densenet161_fl, Resnext101_32x8d_fl, MobilenetV2_fl, Vgg16_bn_fl, Efficientnet_fl
 
 
-font = {
-        'weight' : 'normal',
-        'size'   : 18}
-plt.rc('font',family='Times New Roman') 
-matplotlib.rc('font', **font)
+# font = {
+#         'weight' : 'normal',
+#         'size'   : 18}
+# plt.rc('font',family='Times New Roman') 
+# matplotlib.rc('font', **font)
 
 def test_net(model_fl_1,
             model_fl_2,
@@ -187,10 +187,19 @@ def get_args():
     logging.basicConfig(level=logging.INFO, format='%(levelname)s: %(message)s')
     args = get_args()
 
-    torch.cuda.set_device(args.local_rank)
-    device = torch.device("cuda", args.local_rank)
-
-    #logging.info(f'Using device {device}')
+
+    # Check if CUDA is available
+    if torch.cuda.is_available():
+        logging.info("CUDA is available. Using CUDA...")
+        device = torch.device("cuda",args.local_rank)
+    elif torch.backends.mps.is_available():  # Check if MPS is available (for macOS)
+        logging.info("MPS is available. Using MPS...")
+        device = torch.device("mps")
+    else:
+        logging.info("Neither CUDA nor MPS is available. Using CPU...")
+        device = torch.device("cpu")
+
+    logging.info(f'Using device {device}')
 
     test_dir = args.test_dir
     dataset=args.dataset
@@ -243,31 +252,31 @@ def get_args():
     model_fl_7.to(device=device)
     model_fl_8.to(device=device)
 
-    map_location = {'cuda:%d' % 0: 'cuda:%d' % args.local_rank}
+    #map_location = {'cuda:%d' % 0: 'cuda:%d' % args.local_rank}
     if args.load:
         model_fl_1.load_state_dict(
-            torch.load(checkpoint_path_1, map_location="cuda:0")
+            torch.load(checkpoint_path_1, map_location=device)
         )
         model_fl_2.load_state_dict(
-            torch.load(checkpoint_path_2, map_location="cuda:0")
+            torch.load(checkpoint_path_2, map_location=device)
         )
         model_fl_3.load_state_dict(
-            torch.load(checkpoint_path_3, map_location="cuda:0")
+            torch.load(checkpoint_path_3, map_location=device)
         )
         model_fl_4.load_state_dict(
-            torch.load(checkpoint_path_4, map_location="cuda:0")
+            torch.load(checkpoint_path_4, map_location=device)
         )
         model_fl_5.load_state_dict(
-            torch.load(checkpoint_path_5, map_location="cuda:0")
+            torch.load(checkpoint_path_5, map_location=device)
         )
         model_fl_6.load_state_dict(
-            torch.load(checkpoint_path_6, map_location="cuda:0")
+            torch.load(checkpoint_path_6, map_location=device)
         )
         model_fl_7.load_state_dict(
-            torch.load(checkpoint_path_7, map_location="cuda:0")
+            torch.load(checkpoint_path_7, map_location=device)
         )
         model_fl_8.load_state_dict(
-            torch.load(checkpoint_path_8, map_location="cuda:0")
+            torch.load(checkpoint_path_8, map_location=device)
         )
 
     # faster convolutions, but more memory

M2_Artery_vein/test_outside.py

@@ -280,8 +280,18 @@ def get_args():
 
     logging.basicConfig(level=logging.INFO, format='%(levelname)s: %(message)s')
     args = get_args()
-    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-    #logging.info(f'Using device {device}')
+    # Check if CUDA is available
+    if torch.cuda.is_available():
+        logging.info("CUDA is available. Using CUDA...")
+        device = torch.device("cuda:0")
+    elif torch.backends.mps.is_available():  # Check if MPS is available (for macOS)
+        logging.info("MPS is available. Using MPS...")
+        device = torch.device("mps")
+    else:
+        logging.info("Neither CUDA nor MPS is available. Using CPU...")
+        device = torch.device("cpu")
+
+    logging.info(f'Using device {device}')
 
     img_size = Define_image_size(args.uniform, args.dataset)
     dataset_name = args.dataset
@@ -347,79 +357,79 @@ def get_args():
 
 
     for i in range(1):
-        net_G_1.load_state_dict(torch.load(  checkpoint_saved_1 + 'CP_best_F1_all.pth'))
-        net_G_A_1.load_state_dict(torch.load( checkpoint_saved_1 + 'CP_best_F1_A.pth'))
-        net_G_V_1.load_state_dict(torch.load(checkpoint_saved_1 + 'CP_best_F1_V.pth'))
+        net_G_1.load_state_dict(torch.load(  checkpoint_saved_1 + 'CP_best_F1_all.pth',map_location=device))
+        net_G_A_1.load_state_dict(torch.load( checkpoint_saved_1 + 'CP_best_F1_A.pth',map_location=device))
+        net_G_V_1.load_state_dict(torch.load(checkpoint_saved_1 + 'CP_best_F1_V.pth',map_location=device))
         net_G_1.eval()
         net_G_A_1.eval()
         net_G_V_1.eval()
         net_G_1.to(device=device)
         net_G_A_1.to(device=device)
         net_G_V_1.to(device=device)
 
-        net_G_2.load_state_dict(torch.load(  checkpoint_saved_2 + 'CP_best_F1_all.pth'))
-        net_G_A_2.load_state_dict(torch.load( checkpoint_saved_2 + 'CP_best_F1_A.pth'))
-        net_G_V_2.load_state_dict(torch.load(checkpoint_saved_2 + 'CP_best_F1_V.pth'))
+        net_G_2.load_state_dict(torch.load(  checkpoint_saved_2 + 'CP_best_F1_all.pth',map_location=device))
+        net_G_A_2.load_state_dict(torch.load( checkpoint_saved_2 + 'CP_best_F1_A.pth',map_location=device))
+        net_G_V_2.load_state_dict(torch.load(checkpoint_saved_2 + 'CP_best_F1_V.pth',map_location=device))
         net_G_2.eval()
         net_G_A_2.eval()
         net_G_V_2.eval()
         net_G_2.to(device=device)
         net_G_A_2.to(device=device)
         net_G_V_2.to(device=device)
 
-        net_G_3.load_state_dict(torch.load(  checkpoint_saved_3 + 'CP_best_F1_all.pth'))
-        net_G_A_3.load_state_dict(torch.load( checkpoint_saved_3 + 'CP_best_F1_A.pth'))
-        net_G_V_3.load_state_dict(torch.load(checkpoint_saved_3 + 'CP_best_F1_V.pth'))
+        net_G_3.load_state_dict(torch.load(  checkpoint_saved_3 + 'CP_best_F1_all.pth',map_location=device))
+        net_G_A_3.load_state_dict(torch.load( checkpoint_saved_3 + 'CP_best_F1_A.pth',map_location=device))
+        net_G_V_3.load_state_dict(torch.load(checkpoint_saved_3 + 'CP_best_F1_V.pth',map_location=device))
         net_G_3.eval()
         net_G_A_3.eval()
         net_G_V_3.eval()
         net_G_3.to(device=device)
         net_G_A_3.to(device=device)
         net_G_V_3.to(device=device)
 
-        net_G_4.load_state_dict(torch.load(  checkpoint_saved_4 + 'CP_best_F1_all.pth'))
-        net_G_A_4.load_state_dict(torch.load( checkpoint_saved_4 + 'CP_best_F1_A.pth'))
-        net_G_V_4.load_state_dict(torch.load(checkpoint_saved_4 + 'CP_best_F1_V.pth'))
+        net_G_4.load_state_dict(torch.load(  checkpoint_saved_4 + 'CP_best_F1_all.pth',map_location=device))
+        net_G_A_4.load_state_dict(torch.load( checkpoint_saved_4 + 'CP_best_F1_A.pth',map_location=device))
+        net_G_V_4.load_state_dict(torch.load(checkpoint_saved_4 + 'CP_best_F1_V.pth',map_location=device))
         net_G_4.eval()
         net_G_A_4.eval()
         net_G_V_4.eval()
         net_G_4.to(device=device)
         net_G_A_4.to(device=device)
         net_G_V_4.to(device=device)
 
-        net_G_5.load_state_dict(torch.load(  checkpoint_saved_5 + 'CP_best_F1_all.pth'))
-        net_G_A_5.load_state_dict(torch.load( checkpoint_saved_5 + 'CP_best_F1_A.pth'))
-        net_G_V_5.load_state_dict(torch.load(checkpoint_saved_5 + 'CP_best_F1_V.pth'))
+        net_G_5.load_state_dict(torch.load(  checkpoint_saved_5 + 'CP_best_F1_all.pth',map_location=device))
+        net_G_A_5.load_state_dict(torch.load( checkpoint_saved_5 + 'CP_best_F1_A.pth',map_location=device))
+        net_G_V_5.load_state_dict(torch.load(checkpoint_saved_5 + 'CP_best_F1_V.pth',map_location=device))
         net_G_5.eval()
         net_G_A_5.eval()
         net_G_V_5.eval()
         net_G_5.to(device=device)
         net_G_A_5.to(device=device)
         net_G_V_5.to(device=device)
 
-        net_G_6.load_state_dict(torch.load(  checkpoint_saved_6 + 'CP_best_F1_all.pth'))
-        net_G_A_6.load_state_dict(torch.load( checkpoint_saved_6 + 'CP_best_F1_A.pth'))
-        net_G_V_6.load_state_dict(torch.load(checkpoint_saved_6 + 'CP_best_F1_V.pth'))
+        net_G_6.load_state_dict(torch.load(  checkpoint_saved_6 + 'CP_best_F1_all.pth',map_location=device))
+        net_G_A_6.load_state_dict(torch.load( checkpoint_saved_6 + 'CP_best_F1_A.pth',map_location=device))
+        net_G_V_6.load_state_dict(torch.load(checkpoint_saved_6 + 'CP_best_F1_V.pth',map_location=device))
         net_G_6.eval()
         net_G_A_6.eval()
         net_G_V_6.eval()
         net_G_6.to(device=device)
         net_G_A_6.to(device=device)
         net_G_V_6.to(device=device)
 
-        net_G_7.load_state_dict(torch.load(  checkpoint_saved_7 + 'CP_best_F1_all.pth'))
-        net_G_A_7.load_state_dict(torch.load( checkpoint_saved_7 + 'CP_best_F1_A.pth'))
-        net_G_V_7.load_state_dict(torch.load(checkpoint_saved_7 + 'CP_best_F1_V.pth'))
+        net_G_7.load_state_dict(torch.load(  checkpoint_saved_7 + 'CP_best_F1_all.pth',map_location=device))
+        net_G_A_7.load_state_dict(torch.load( checkpoint_saved_7 + 'CP_best_F1_A.pth',map_location=device))
+        net_G_V_7.load_state_dict(torch.load(checkpoint_saved_7 + 'CP_best_F1_V.pth',map_location=device))
         net_G_7.eval()
         net_G_A_7.eval()
         net_G_V_7.eval()
         net_G_7.to(device=device)
         net_G_A_7.to(device=device)
         net_G_V_7.to(device=device)
 
-        net_G_8.load_state_dict(torch.load(  checkpoint_saved_8 + 'CP_best_F1_all.pth'))
-        net_G_A_8.load_state_dict(torch.load( checkpoint_saved_8 + 'CP_best_F1_A.pth'))
-        net_G_V_8.load_state_dict(torch.load(checkpoint_saved_8 + 'CP_best_F1_V.pth'))
+        net_G_8.load_state_dict(torch.load(  checkpoint_saved_8 + 'CP_best_F1_all.pth',map_location=device))
+        net_G_A_8.load_state_dict(torch.load( checkpoint_saved_8 + 'CP_best_F1_A.pth',map_location=device))
+        net_G_V_8.load_state_dict(torch.load(checkpoint_saved_8 + 'CP_best_F1_V.pth',map_location=device))
         net_G_8.eval()
         net_G_A_8.eval()
         net_G_V_8.eval()

M2_Vessel_seg/test_outside_integrated.py

@@ -203,34 +203,34 @@ def test_net(data_path, batch_size, device, dataset_train, dataset_test, image_s
     net_10 = Segmenter(input_channels=3, n_filters = 32, n_classes=1, bilinear=False)
 
 
-    net_1.load_state_dict(torch.load(dir_checkpoint_1 + 'G_best_F1_epoch.pth'))
+    net_1.load_state_dict(torch.load(dir_checkpoint_1 + 'G_best_F1_epoch.pth',map_location=device))
     net_1.eval()
     net_1.to(device=device)
-    net_2.load_state_dict(torch.load(dir_checkpoint_2 + 'G_best_F1_epoch.pth'))
+    net_2.load_state_dict(torch.load(dir_checkpoint_2 + 'G_best_F1_epoch.pth',map_location=device))
     net_2.eval()
     net_2.to(device=device)
-    net_3.load_state_dict(torch.load(dir_checkpoint_3 + 'G_best_F1_epoch.pth'))
+    net_3.load_state_dict(torch.load(dir_checkpoint_3 + 'G_best_F1_epoch.pth',map_location=device))
     net_3.eval()
     net_3.to(device=device)
-    net_4.load_state_dict(torch.load(dir_checkpoint_4 + 'G_best_F1_epoch.pth'))
+    net_4.load_state_dict(torch.load(dir_checkpoint_4 + 'G_best_F1_epoch.pth',map_location=device))
     net_4.eval()
     net_4.to(device=device)
-    net_5.load_state_dict(torch.load(dir_checkpoint_5 + 'G_best_F1_epoch.pth'))
+    net_5.load_state_dict(torch.load(dir_checkpoint_5 + 'G_best_F1_epoch.pth',map_location=device))
     net_5.eval()
     net_5.to(device=device)
-    net_6.load_state_dict(torch.load(dir_checkpoint_6 + 'G_best_F1_epoch.pth'))
+    net_6.load_state_dict(torch.load(dir_checkpoint_6 + 'G_best_F1_epoch.pth',map_location=device))
     net_6.eval()
     net_6.to(device=device)
-    net_7.load_state_dict(torch.load(dir_checkpoint_7 + 'G_best_F1_epoch.pth'))
+    net_7.load_state_dict(torch.load(dir_checkpoint_7 + 'G_best_F1_epoch.pth',map_location=device))
     net_7.eval()
     net_7.to(device=device)
-    net_8.load_state_dict(torch.load(dir_checkpoint_8 + 'G_best_F1_epoch.pth'))
+    net_8.load_state_dict(torch.load(dir_checkpoint_8 + 'G_best_F1_epoch.pth',map_location=device))
     net_8.eval()
     net_8.to(device=device)
-    net_9.load_state_dict(torch.load(dir_checkpoint_9 + 'G_best_F1_epoch.pth'))
+    net_9.load_state_dict(torch.load(dir_checkpoint_9 + 'G_best_F1_epoch.pth',map_location=device))
     net_9.eval()
     net_9.to(device=device)
-    net_10.load_state_dict(torch.load(dir_checkpoint_10 + 'G_best_F1_epoch.pth'))
+    net_10.load_state_dict(torch.load(dir_checkpoint_10 + 'G_best_F1_epoch.pth',map_location=device))
     net_10.eval()
     net_10.to(device=device)
 
@@ -285,9 +285,20 @@ def get_args():
 
     logging.basicConfig(level=logging.INFO, format='%(levelname)s: %(message)s')
     args = get_args()
-    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+    # Check if CUDA is available
+    if torch.cuda.is_available():
+        logging.info("CUDA is available. Using CUDA...")
+        device = torch.device("cuda:0")
+    elif torch.backends.mps.is_available():  # Check if MPS is available (for macOS)
+        logging.info("MPS is available. Using MPS...")
+        device = torch.device("mps")
+    else:
+        logging.info("Neither CUDA nor MPS is available. Using CPU...")
+        device = torch.device("cpu")
+
     logging.info(f'Using device {device}')
 
+
     image_size = Define_image_size(args.uniform, args.dataset)
     lr = args.lr